Figure 1. Adam Lewis and Dave Lowe, refuelling with a Queensland Boating Patrol team at Banana Cay, 190 kilometres from land and only exposed at low tide
The first comprehensive mapping of the Great Barrier Reef was completed in the 1980s using Landsat multispectral data and miscellaneous small-scale aerial photography, but with limited ground control. The resulting maps, generally suitable for interpretation at scales of 1:250,000, are known to include significant errors due to the lack of ground control. With increasing use of the region for shipping, tourism, recreation, commercial fishing and other activities, more detailed and accurate maps became a priority of the Great Barrier Reef Marine Park Authority.
Landsat-7, with its reduced costs, opened a window of opportunity -- to re-map the park more accurately and at higher resolution. Logistical problems lay ahead, though. How does one collect the large number of ground control points required to accurately register 25 Landsat scenes over the ocean? Reefs and cays, despite being numerous, only cover 6 per cent of the region and are widely dispersed. Use of ships would have made for a long holiday, but would have required more time and money than we could afford.
On land, many areas adjacent to the marine park are remote and seasonally inaccessible, making use of vehicles difficult or impossible. We found the solution in helicopters. With just enough range to reach some of the most remote reefs in the park (250 km offshore), the Robinson R44 became the workhorse of this part of the project. Despite high hourly rates, the speed of travel made this an extremely cost-effective method of accessing ground control points (GCPs).
White Haven beach, Whitsunday Islands. (Alas, no time to stop!)
Finding suitable GCPs over the Great Barrier Reef posed another set of problems. Highly reflective and contrasting features, such as sand cays, are unreliable as control points because of tidal variations and the influence of directional wave breaks. Sand cays can also move seasonally, just as tropical vegetation can change seasonally to mask points on land.
In contrast, coral reef bombora, or 'bommies', are wonderfully stable over time but have low albedo and are often submerged. Furthermore, on some reefs the bommies are too small to be detected on Landsat imagery, or are absent altogether, or are too large to be ideal control points. Planning the ground control missions became a matter of finding the right type of bommie in the right part of the image, while juggling logistical considerations -- passenger weights, headwinds, tides, distances off-shore, fuel supply rendezvous, sunset times, and coordination with the road crew, to name a few. Each ground control mission had to be carefully planned, and we usually had only minutes to spare.
Inside the R44. GPS receiver with spare differential correction receiver.
We optimised our ground control missions by targeting clearly visible reefs that lay at the corners and sides of images, and maximising our use of some 300 pre-existing orthophoto mosaics (developed by the Australian Survey and Land Information Group in the 1980s). This ensured that the geometry of the control points was close to optimum and that each control point was used as often as possible -- sometimes contributing to the rectification of three adjacent images. Trip preparation included local image enhancement and printouts of images over each control site.
We found that the actual control points on a reef could not be pre-planned. Instead, from an altitude of about 200 metres, on-the-spot judgement calls were necessary to orientate the image-map to the real world, identify suitable reef features, and choose the best set of GCPs for that site.
Co-ordinates were collected using a Garmin-X12 GPS with differential correction to the Australian Maritime Safety Authority beacon network, while hovering below a metre. For economy and safety, we held position only long enough for the GPS readings to stabilise; the GCP was marked precisely on the image-map at that time. Generally we collected at least three GCPs per reef. We collected 423 points with differential GPS and added to this 237 points sourced from 1:3500--1:10,000 orthophoto maps.
Images were rectified avoiding any land-based control points that would involve significant terrain displacement. We used a preliminary model to flag and remove outliers arising through GPS failure or operator error. We then aggregated closely associated control points (e.g: points from over a single reef) into a mean control point before completing the image rectification.
The aggregation was expected to reduce the noise associated with the GCP field methods, and to avoid bias toward parts of the image where, fortuitously, we had more control points. Ultimately, however, this averaging process had little impact on results.
A field of very large bommies within a reef lagoon
Our attention to detail in the GCP planning paid off in the rectification results. Up to 40 (average of 14) checkpoints per image, distributed across the images and away from control points, were withheld from the rectification. The checkpoints give a conservative estimate of the 50 and 90 percentile error as 14.9 and 31.1 metres respectively, with average error under 20 metres.
The rectification was sufficiently precise to suggest problems with some of underlying image geometry; in several cases images could not be rectified without 40+ metre RMS errors. Replacement images, one volunteered from the Australian Centre for Remote Sensing, resolved the problem, suggesting that the systematic image rectification may in some cases be incomplete.
The reef mapping project is already delivering benefits. A colour-balanced image mosaic of the Great Barrier Reef will support management and planning as the Authority implements its plans for more green zones in the marine park, and accurate images are providing strong evidence in the prosecution of illegal fishing. For the first time, managers have at their fingertips 'the real thing', with known accuracy. The classification stage of the project, now under way, will lead to far more detailed maps of coral reefs.
